Deep learning-based fringe pattern transformation method for phase calculation

Fringe projection profilometry (i.e., FPP) has been one of the most popular techniques in three-dimensional (i.e., 3-D) measurement. In FPP, it is necessary to obtain accurate desired phase by using a small number of fringes in dynamic measurement. Recently, fringe pattern transformation method (i.e., FPTM) is proposed based on deep learning, which can achieve accurate 3-D measurement using a single fringe, but the phase error is still higher than the phase-shifting algorithm. In this paper, the phase error of FPTM is analyzed and the relationship between it and local depth change rate is illustrated firstly. Then, the accuracy of FPTM can be improved by using more fringes. Compared with traditional methods, FPTM can achieve higher precision 3-D measurement when less fringes are used.